The clearing house has called far more initial margin than your own risk system predicted. How do you find out why before the deadline?

Pay first, explain second

Begin by separating two problems that share a deadline. The call is a contractual obligation under the clearing house's rulebook, and a missed call is a default event regardless of whether your analysis eventually shows the number was too high. So the funding decision is made on the assumption that the call is correct, and the investigation runs in parallel. A candidate who describes disputing the figure before arranging the cash has inverted the priority, and the follow-up question is always what happens at the cut-off.

That framing also tells you what the investigation is for. You are not judging the clearing house's model. You are reproducing its output, because the only thing you can act on is a difference you can point at.

Reconcile the population before you touch the model

The overwhelming majority of large, sudden margin surprises are not modelling differences. They are differences in what is being margined, and they are found in minutes rather than hours if you look there first.

Compare the clearing house's position report line by line against your own view of the same account. The usual causes are mundane. A trade given up to you that you have not booked, or booked to the wrong account. An allocation that has not been through, so a block sits in a suffix account that margins on its own. A sign error, which is expensive because a short shown as a long removes an offset and adds an equal exposure in the opposite direction. An expired or exercised contract still open in your book. A contract multiplier or lot size taken from stale reference data, which scales the whole position. And an account structure mismatch, where positions you assume are in one netting set are recorded in two.

Only when the two position sets agree, at instrument and account level, is a difference in margin a difference about risk. Reversing that order is how a team spends a day investigating volatility parameters for a break that was a missing give-up.

Attribute the remainder in layers

With the population agreed, work down through the inputs in an order that stops as soon as it explains the gap.

Attribution of a margin difference, most likely cause first

  Population      positions, accounts, multipliers, expiries
  Prices          the clearing house's official settlement prices,
                  and its FX rates, not your intraday marks
  Parameters      scan ranges, volatility inputs and correlations as
                  published for today, including overnight changes
  Netting scope   which positions are permitted to offset which
  Add-ons         concentration, liquidity, wrong-way and credit
  Floors          minimum charges and anti-procyclicality measures

Prices come first among the inputs because they are the easiest to get wrong quietly. A clearing house margins from its own official settlement prices and its own conversion rates, struck at its own time. Your risk system probably uses a market close and a rate from a different source at a different moment, and on a volatile day those two are not the same number.

Parameters come next, and this is where a call jumps overnight with no change to your book at all. Margin models are recalibrated, and a clearing house that widens a scan range or raises a volatility input after a turbulent session will raise every member's requirement simultaneously. Because regulation requires those models to include anti-procyclicality measures, such as giving weight to a stressed historical period so requirements do not collapse in calm markets, a stress observation entering or leaving the look-back window can move the number sharply. Checking the clearing house's published parameter notices for the day is a five-minute test that resolves a surprising number of these.

Netting scope is the structural reason two correct numbers differ

Your internal risk system almost certainly computes exposure across the whole portfolio, letting anything that offsets anything else do so. A clearing house does not, and its constraints are legal rather than statistical.

Offsets are permitted only within a defined scope, so positions in different product groups or different clearing services can be economically opposed and still attract margin on both. Where the clearing house margins client positions gross rather than netting them across clients, two clients of yours holding opposite positions produce two requirements instead of none, and the total bears no resemblance to a portfolio figure. Segregation between house and client accounts has the same effect by design, because client collateral is not available to cover house exposure.

Then the add-ons, which are charges your own model has no reason to produce. A concentration charge because your position is large relative to the market's liquidity. A liquidity or close-out charge reflecting what it would cost to unwind rather than what the position is worth. A wrong-way charge where your collateral or your credit is correlated with the exposure. These are deliberately conservative and are not errors to be argued away.

There is also a horizon effect worth being precise about. Initial margin is sized over an assumed close-out period, the margin period of risk, whereas an internal figure is often a one-day measure. Scaling one day to two under the standard assumption of independent, identically distributed returns multiplies by the square root of two, roughly 1.41. So a portfolio your system marks at ten million of one-day risk is around fourteen million on a two-day horizon before any add-on. State the assumption when you use it, because the reason the real difference exceeds that factor is that returns in a stressed market are neither independent nor identically distributed, which is exactly what the add-ons and the stressed look-back are compensating for.

The number you needed was a forecast, not an explanation

An investigation that concludes at 16:00 has answered a question the treasurer needed answered at 09:00. So the durable fix is to stop treating margin as an incoming figure and start treating it as a quantity you compute yourself.

That means implementing the clearing house's published methodology, or using the calculator it provides, against your own positions, on the clearing house's own parameters and prices. Run it as a daily control with a tolerance and an owner, so a small unexplained difference is visible while it is still small and the day a large one appears you already know the previous day tied. Run it again as a what-if service the desk can call before it trades, because the margin consequence of a hedge is frequently larger than its market-risk benefit and a trader who cannot see that will place trades the treasury has to fund. And keep the attribution automated, so a break arrives with its layers already computed rather than as a total to be investigated.

Where candidates lose the argument

The weak answer treats the difference as evidence that the clearing house is wrong, and offers to send its own risk numbers as a challenge. That misreads what margin is. It is not an opinion about your risk that you and the clearing house are jointly forming; it is a term of the contract you entered when you cleared, produced by a documented model on published parameters, and the clearing house has no obligation to agree with your view of the same portfolio. Your obligation is to reproduce it, to fund it, and to have the desk see it in advance.

Reconcile the positions before you reconcile the model, and build the capability to compute the call yourself. A margin figure you can only explain after it arrives is a funding surprise every time the market moves.